[acolomba]

Rework the behave suite so it follows
skills/behave-behavioral-testing/SKILL.md, and fix the exit status of the
Docker image. Each review and each fix of the suite ran in a separate agent.
The loop stopped when a fresh review found no major problems.

With RUN_ONCE set, entrypoint.sh always exited 0, because the script
ended with an if statement without else. A failed sync looked like a
success to whatever started the container, and in docker mode no behave
check of the exit code could fail. The script now keeps the sync status and
exits with it. Cron mode is unchanged: crond starts after the first sync,
whatever its result.

Several checks could pass without testing anything. Three destination checks
returned early when a setup step was missing, and most checks tested only a
subset of the files. All checks now compare the whole set of files in the
destination, or the whole set of requests that the fake dashcam received. A
failed check prints the missing and the unexpected files. The exit code check
prints the output of blackvuesync.

Step modules are now grouped by function: blackvuesync runs, the dashcam, the
destination, and general checks. One step, blackvuesync runs with ...,
replaces six. The Gherkin uses the active voice, and each feature has a short
description.

The fake dashcam listens on a random port on 127.0.0.1 and shuts down at the
end of the run. It records the files that blackvuesync requests. A scenario
can make it fail a download or the recording list. Cleanup goes through
context.add_cleanup. When a scenario fails, the run keeps only the directory
of that scenario, with its destination and a log of each blackvuesync run.

New scenarios cover a failed recording list (exit code 2), daily and weekly
grouping, leftover partial downloads, and a destination that already has some
of the dashcam recordings. A step that backdates the failure markers replaces
a 2-second sleep.

The suite now requires behave 1.3.3, the latest release. It uses tag
expressions v2 (not @ignore) and active tags (@use.with_protocol=legacy)
in place of the hand-written skips. Fixtures, the Rule keyword, and
cucumber expressions did not make this suite simpler, so the suite does not
use them. mypy now ignores the missing behave stubs, so local runs match the
pre-commit hook.

The unit-test and behave guidelines now prefer PyHamcrest's assert_that()
with a matcher to state the condition of an assertion. The behave steps use
equal_to() for every whole-value comparison, and pass the set differences
and the output of blackvuesync as the reason. The change also removes the
CLAUDE.md files that only imported AGENTS.md.

Validation: scripts/check.sh passed, with 71.51% unit test coverage. The
behave results are 27 passed and 1 skipped for the current protocol, and 28
passed for the legacy protocol. Docker mode gives the same results, and the
listing error scenario runs there too. The coverage runs passed. A direct run
on Python 3.9 passed. All pre-commit hooks passed. The package version
remains 2.2.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XhzDTTf8Wj1sNRHqJdHcUz

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud